powerpc/mm: Fix potential access to freed pages when using hugetlbfs
authorBenjamin Herrenschmidt <benh@kernel.crashing.org>
Mon, 15 Jun 2009 16:53:43 +0000 (16:53 +0000)
committerBenjamin Herrenschmidt <benh@kernel.crashing.org>
Fri, 26 Jun 2009 06:55:36 +0000 (16:55 +1000)
commit6c16a74d423f584ed80815ee7b944f5b578dd37a
treec8ef820f92601b01ed5b90e840ed5123e51dfa0f
parentf694cda89250f38782a5fd0b2d32fe33a39dcf37
powerpc/mm: Fix potential access to freed pages when using hugetlbfs

When using 64k page sizes, our PTE pages are split in two halves,
the second half containing the "extension" used to keep track of
individual 4k pages when not using HW 64k pages.

However, our page tables used for hugetlb have a slightly different
format and don't carry that "second half".

Our code that batched PTEs to be invalidated unconditionally reads
the "second half" (to put it into the batch), which means that when
called to invalidate hugetlb PTEs, it will access unrelated memory.

It breaks when CONFIG_DEBUG_PAGEALLOC is enabled.

This fixes it by only accessing the second half when the _PAGE_COMBO
bit is set in the first half, which indicates that we are dealing with
a "combo" page which represents 16x4k subpages. Anything else shouldn't
have this bit set and thus not require loading from the second half.

Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
arch/powerpc/include/asm/pte-hash64-64k.h